Some characterizations of the class Em(Ω) and applications

ANNALES POLONICI MATHEMATICI 115.2 (2015) Some characterizations of the class Em(Ω) and applications by Hai Mau Le (Hanoi), Hong Xuan Nguyen (Hanoi) and Hung Viet Vu (Son La) Abstract.We give some characterizations of the class Em(Ω) and use them to establish a lower estimate for the log canonical threshold of plurisubharmonic functions in this class. 1. Introduction. The complex Monge–Ampère operator has a central role in pluripotential theory and has been extensively studied for many years. This operator was used to obtain many important results of pluripoten- tial theory in Cn, n > 1. An example of such application is the proof of quasi-continuity of plurisubharmonic functions, yielding the pluripolarity of negligible sets. In [BT1] Bedford and Taylor have shown that this operator is well defined on the class of locally bounded plurisubharmonic functions with range in the class of nonnegative measures. Recently, to extend the domain of definition of this operator to plurisubharmonic functions which may or not be locally bounded, Cegrell [C1, C2] has introduced and investigated the classes E0(Ω), F(Ω) and E(Ω) on which the complex Monge–Ampère oper- ator is well defined. He has developed pluripotential theory on these classes. To extend the class of plurisubharmonic functions and to study a class of complex differential operators more general than the Monge–Ampère oper- ator, in [B1] and [DK2], the authors introduced m-subharmonic functions and studied the complex Hessian operator. They were also interested in the complex Hessian equations in Cn and on compact Kähler manifolds. In order to continue the study of the complex Hessian operator for m-subharmonic functions which are not locally bounded, in a recent preprint [Lu], Chinh Hoang Lu introduced the Cegrell classes E0m(Ω), Fm(Ω) and Em(Ω) asso- ciated to m-subharmonic functions, and proved that the complex Hessian operator is well defined on these classes. Thus it is of interest to obtain a characterization of these classes analytically and geometrically. 2010 Mathematics Subject Classification: 32U05, 32U15, 32U40, 32W20. Key words and phrases: m-subharmonic functions, complex m-Hessian operator, charac- terizations of Em(Ω), log canonical threshold. DOI: 10.4064/ap115-2-3 [145] c© Instytut Matematyczny PAN, 2015 146 H. M. Le et al. In Section 3 we show the local property and give an analytic charac- terization of the class Em(Ω). At the beginning of Section 4, we prove the following: Theorem 4.1. Let Ω be a bounded hyperconvex domain in Cn and u ∈ Em(Ω) ∩ PSH−(Ω). Then for all c > 0, the upper level sets E(u, c) = {z ∈ Ω : ν(u, z) ≥ c} are analytic subsets in Ω of dimension ≤ n−m. Next, by relying on a recent result of Demailly and Pham [DP] we give a lower bound for the log canonical threshold of plurisubharmonic functions in the class Em(Ω). This is the following theorem. Theorem 4.5. Let u ∈ PSH(Ω) ∩ Em(Ω), 1 ≤ m ≤ n − 1 and 0 ∈ Ω. Then cu(0) ≥ m∑ j=1 ej−1(u) ej(u) , where e0(u) = 1. Note that in the above theorem cu(0) and ej(u) denote, respectively, the log canonical threshold and the intersection number of the plurisubharmonic function u, whose definitions are given in Section 4. In the case m = n, from the above theorem we get the result of Demailly and Pham. Finally, using a result in [FS] we will prove the same lower bound for the log canonical threshold of plurisubharmonic functions which are bounded outside a closed subset of small Hausdorff measure: Theorem 4.6. Let Ω be an open subset in Cn, 0 ∈ Ω and E ⊂ Ω be a closed subset in Ω with H2(n−m)+2(E) = 0, where 1 ≤ m ≤ n − 1. Assume that u ∈ PSH(Ω) ∩ L∞(Ω \ E). Then cu(0) ≥ m∑ j=1 ej−1(u) ej(u) , where e0(u) = 1. The paper is organized as follows. In Section 2 we recall the definitions and results concerningm-subharmonic functions, which were introduced and investigated intensively in recent years by many authors (see [B1], [DK2]). We also recall the Cegrell classes ofm-subharmonic functions: E0m(Ω), Fm(Ω) and Em(Ω) introduced and studied in [Lu]. At the same time, we deal with the Lelong numbers associated to a closed positive current T and to a plurisubharmonic function ϕ on an open set Ω ⊂ Cn. Results of Siu and Demailly on the analyticity of upper level sets of the Lelong numbers asso- ciated to a closed positive current T are recalled in that section. Section 3 is Some characterizations of the class Em(Ω) 147 devoted to the proof of the local property and of an analytic characteriza- tion of Em(Ω). Next, in Section 4 we give a geometrical characterization of Em(Ω), and by relying on the result of Demailly and Pham [DP], we give a lower estimate for the log canonical threshold of plurisubharmonic functions in the class Em(Ω) and of plurisubharmonic functions which are bounded outside a subset of small Hausdorff measure. 2. Preliminaries. Some elements of pluripotential theory that will be used throughout the paper can be found in [BT1], [K1], [K2], [K3], while elements of the theory of m-subharmonic functions and the complex Hessian operator can be found in [B1], [DK2], [SA]. Now we recall the definition of the class of m-subharmonic functions introduced by Błocki [B1] and the classes E0m(Ω) and Fm(Ω) introduced and investigated by Chinh Hoang Lu [Lu]. Let Ω be an open subset in Cn. We denote by PSH(Ω) the set of plurisubharmonic functions on Ω, while PSH−(Ω) denotes the set of negative plurisubharmonic functions on Ω. By β = ddc|z|2 we denote the canonical Kähler form on Cn with the volume form dVn = 1n!β n where d = ∂ + ∂ and dc = ∂−∂4i , hence dd c = i2∂∂. 2.1. First, we recall the class of m-subharmonic functions introduced and investigated in [B1]. For 1 ≤ m ≤ n, we define Γ̂m = {η ∈ C(1,1) : η ∧ βn−1 ≥ 0, . . . , ηm ∧ βn−m ≥ 0}, where C(1,1) denotes the space of (1, 1)-forms with constant coefficients. Definition 2.1. Let u be a subharmonic function on an open subset Ω ⊂ Cn. Then u is said to be an m-subharmonic function on Ω if for every η1, . . . , ηm−1 in Γ̂m the inequality ddcu ∧ η1 ∧ · · · ∧ ηm−1 ∧ βn−m ≥ 0 holds in the sense of currents. By SHm(Ω) we denote the set of allm-subharmonic functions on Ω, while SH−m(Ω) denotes the set of negative m-subharmonic functions on Ω. Now as- sume that Ω is an open set in Cn and u ∈ C2(Ω). Then from [B1, Proposition 3.1] (see also [SA, Definition 1.2]) we note that u is m-subharmonic on Ω if and only if (ddcu)k ∧ βn−k ≥ 0 for k = 1, . . . ,m. Now as in [B1] and [DK2], we define the complex Hessian operator of locally bounded m-subharmonic functions. Definition 2.2. Assume that u1, . . . , up ∈ SHm(Ω)∩L∞loc(Ω). Then the complex Hessian operator Hm(u1, . . . , up) is defined inductively by ddcup ∧ · · · ∧ ddcu1 ∧ βn−m = ddc(upddcup−1 ∧ · · · ∧ ddcu1 ∧ βn−m). 148 H. M. Le et al. In [B1] and [DK2] it is proved that Hm(u1, . . . , up) is a closed positive current of bidegree (n −m + p, n −m + p) and this operator is continuous under decreasing sequences of locally bounded m-subharmonic functions. In particular, when u = u1 = · · · = um ∈ SHm(Ω)∩L∞loc(Ω), the Borel measure Hm(u) = (dd cu)m ∧ βn−m is well defined and is called the complex Hessian of u. Example 2.3. By using an example due to Sadullaev and Abullaev [SA] we show that there exists a function which ism-subharmonic but not (m+1)- subharmonic. Let Ω ⊂ Cn be a domain and 0 /∈ Ω. Consider the Riesz kernel Km(z) = − 1|z|2(n/m−1) , 1 ≤ m < n. We note that Km ∈ C2(Ω). As in [SA] we have (ddcKm) k ∧ βn−k = n(n/m− 1)k(1− k/m)|z|−2kn/mβn. Thus (ddcKm)k ∧ βn−k ≥ 0 for all k = 1, . . . ,m, and hence Km ∈ SHm(Ω). However, (ddcKm)m+1 ∧ βn−m−1 < 0 and so Km /∈ SHm+1(Ω). 2.2. Now we recall the definition of m-maximal subharmonic functions introduced and investigated in [B1]. Definition 2.4. An m-subharmonic function u ∈ SHm(Ω) is called m- maximal if for every K b Ω and every v ∈ SHm(Ω), if v ≤ u on Ω \K then v ≤ u on Ω. We denote by MSHm(Ω) the set of m-maximal functions on Ω. Theorem 3.6 in [B1] states that a locally bounded m-subharmonic function u on a bounded domain Ω ⊂ Cn belongs to MSHm(Ω) if and only if it solves the homogeneous Hessian equation Hm(u) = (ddcu)m ∧ βn−m = 0. 2.3. Next, we recall the classes E0m(Ω), Fm(Ω) and Em(Ω) introduced and investigated in [Lu]. Let Ω be a bounded hyperconvex domain in Cn. Set E0m = E0m(Ω) = { u ∈ SH−m(Ω) ∩ L∞(Ω) : lim z→∂Ω u(z) = 0,  Ω Hm(u) <∞ } , Fm = Fm(Ω) = { u ∈ SH−m(Ω) : ∃ E0m 3 uj ↘ u, sup j  Ω Hm(uj) <∞ } , Em = Em(Ω) = { u ∈ SH−m(Ω) : ∀z0 ∈ Ω, ∃ a neighborhood ω 3 z0, and ∃ E0m 3 uj ↘ u on ω, sup j  Ω Hm(uj) <∞ } . In the casem = n these classes coincide with, respectively, the classes E0(Ω), F(Ω) and E(Ω) introduced and investigated earlier by Cegrell [C2]. Some characterizations of the class Em(Ω) 149 From [Lu, Theorem 3.14] it follows that if u ∈ Em(Ω), the complex Hes- sian Hm(u) = (ddcu)m∧βn−m is well defined and is a Radon measure on Ω. On the other hand, by [Lu, Remark 3.6] the following description of Em(Ω) may be given: Em = Em(Ω) = {u ∈ SH−m(Ω) : ∀U b Ω, ∃v ∈ Fm(Ω), v = u on U}. 2.4. We recall the definition of the pluricomplex Green function. Let Ω be an open subset in Cn, and let a be a point in Ω. The pluricomplex Green function with pole at a, denoted by ga, is defined by ga(z) = sup{u(z) : u ∈ PSH−(Ω), u(z) ≤ log ‖z − a‖+ cu for z near a}. It is well known that if Ω is bounded and B(a, r) ⊂ Ω ⊂ B(a,R) then [K1, Proposition 6.1.1] implies that log(‖z − a‖/R) ≤ ga(z) ≤ log(‖z − a‖/r) for z ∈ Ω, and z 7→ ga(z) is a negative plurisubharmonic function with a logarithmic pole at a. In the case when Ω is a bounded hyperconvex domain we have limz→∂Ω ga(z) = 0. At the same time, by a result of Demailly [D1], the Monge–Ampère measure (ddcga)n is well defined and (ddcga)n = δa, where δa is the Dirac measure at a. On the other hand, it is not difficult to see that ga ∈ F(Ω). 2.5. Now we recall the definition of Lelong numbers associated to a closed positive current T and Lelong numbers of a plurisubharmonic function in- troduced and investigated in [D2] and [D3]. Let Ω ⊂ Cn be an open set and T be a closed positive current of bidimension (p, p) on Ω. Assume that ϕ is a plurisubharmonic function bounded near the boundary ∂Ω of Ω. Then as in [D3] the measure T ∧ (ddcϕ)p is well defined on Ω. The Lelong number of T with respect to the weight ϕ is denoted by ν(T, ϕ) and defined by ν(T, ϕ) =  {ϕ=−∞} T ∧ (ddcϕ)p = lim r→−∞  {ϕ<r} T ∧ (ddcϕ)p. If a ∈ Ω and we take ϕa(z) = log ‖z − a‖ then we get the definition of the Lelong number of T at a which we denote by ν(T, a). Thus ν(T, a) =  {a} T ∧ (ddcϕa)p = lim r→0  {‖z−a‖<r} T ∧ (ddcϕa)p. By 2.4 and by using the comparison theorems for Lelong numbers in [D3] we note that ν(T, a) can also be defined by ν(T, a) =  {a} T ∧ (ddcga)p = lim r→0  {‖z−a‖<r} T ∧ (ddcga)p. Now assume that a ∈ Ω and ϕ ∈ PSH−(Ω). If we take T = ddcϕ then we have the definition of the Lelong number of ϕ at a. Namely, with the 150 H. M. Le et al. notation ϕa above, the Lelong number of ϕ at a, denoted by ν(ϕ, a), is defined by ν(ϕ, a) =  {a} ddcϕ ∧ (ddcϕa)n−1 = lim r→0  {‖z−a‖<r} ddcϕ ∧ (ddcϕa)n−1. As above, we can also define ν(ϕ, a) by ν(ϕ, a) =  {a} ddcϕ ∧ (ddcga)n−1 = lim r→0  {‖z−a‖<r} ddcϕ ∧ (ddcga)n−1. A celebrated result of Siu [Si] for upper level sets of the Lelong numbers of plurisubharmonic functions, later generalized by Demailly [D3] to the Lelong numbers of closed positive currents, says that if T is a closed positive current of bidimension (p, p) on an open set Ω ⊂ Cn then for all c > 0 the upper level sets Ec(T ) = {x ∈ Ω : ν(T, x) ≥ c} are analytic subsets of Ω of dimension ≤ p. 2.6. Throughout the paper we write A . B if there exists a constant C such that A ≤ CB. 3. The local property and an analytic characterization for the class Em(Ω). In this section we show that to belong to the class Em(Ω) is a local property. Relying on this result we give an analytic characterization for this class. First we need the following. Lemma 3.1. Let u, v ∈ SH−m(Ω) ∩ L∞(Ω) with u ≤ v on Ω and T = ddcϕ1∧· · ·∧ddcϕm−1∧βn−m with ϕj ∈ SH−m(Ω)∩L∞(Ω), j = 1, . . . ,m−1. Then for every p ≥ 0 we have Ω′ (−u)pddcv ∧ T ≤ c  Ω′′ (−u)p(ddcu+ |u|β) ∧ T, where Ω′ b Ω′′ b Ω and c is a constant depending on Ω′, Ω′′, Ω and p. Proof. Repeat the argument for [LPH, Lemma 3.2]. We also need the following result on subextension for the class Fm(Ω). Lemma 3.2. Assume that Ω b Ω˜ and u ∈ Fm(Ω). Then there exists a u˜ ∈ Fm(Ω˜) such that u˜ ≤ u on Ω. Proof. We split the proof into three steps. Step 1. We prove that if v ∈ C(Ω˜), v ≤ 0, supp v b Ω˜ then v˜ := sup{w ∈ SH−m(Ω˜) : w ≤ v on Ω˜} ∈ E0m(Ω˜) ∩ C(Ω˜) and (ddcv˜)m ∧ βn−m = 0 on {v˜ < v}. Indeed, let ϕ ∈ E0m(Ω˜) ∩ C(Ω˜) with ϕ ≤ infΩ˜ v on supp v. Since ϕ ≤ v˜ we get v˜ ∈ E0m(Ω˜). Moreover, by [B1, Proposition 3.2] we have Some characterizations of the class Em(Ω) 151 v˜ ∈ C(Ω˜). Let w ∈ SH−m({v˜ < v}) be such that w ≤ v˜ outside a compact subset K of {v˜ < v}. Define w1 = { max(w, v˜) on {v˜ < v}, v˜ on Ω˜ \ {v˜ < v}. Since v˜ and v are continuous, ε = − supK(v˜− v) > 0. Choose δ ∈ (0, 1) such that −δ inf Ω˜ v˜ < ε. Then (1−δ)v˜ ≤ v˜+ε ≤ v onK. Hence, (1−δ)v˜+δw1 ≤ v on Ω˜, and we get (1 − δ)v˜ + δw1 ≤ v˜. Thus, w ≤ v˜ on {v˜ < v}. Therefore, v˜ is m-maximal in {v˜ < v}. By [B1] we get (ddcv˜)m ∧ βn−m = 0 on {v˜ < v}. Step 2. Next, we prove that if u ∈ E0m(Ω) ∩ C(Ω) then there exists u˜ ∈ E0m(Ω˜) for which (ddcu˜)m ∧ βn−m = 0 on (Ω˜\Ω) ∪ ({u˜ < u} ∩ Ω) and (ddcu˜)m ∧ βn−m ≤ (ddcu)m ∧ βn−m on {u˜ = u} ∩Ω. Indeed, set v = { u on Ω, 0 on Ω˜ \Ω. It is easy to see that v ∈ C(Ω) and supp v ⊂ Ω b Ω˜. Hence, by Step 1 we have u˜ = v˜ ∈ E0m(Ω˜) ∩ C(Ω˜) and (ddcu˜)m ∧ βn−m = 0 on {v˜ < v} = (Ω˜ \Ω) ∪ ({u˜ < u} ∩Ω). Let K be a compact set in {u˜ = u} ∩Ω. Then for ε > 0 we have K b {u˜+ ε > u} ∩Ω, and so K (ddcu˜)m ∧ βn−m =  K 1{u˜+ε>u}(ddcu˜)m ∧ βn−m =  K 1{u˜+ε>u}(ddc max(u˜+ ε, u))m ∧ βn−m ≤  K (ddc max(u˜+ ε, u))m ∧ βn−m, where the equality in the second line follows as in [BT2] (see also [Lu, proof of Theorem 3.23]). However, max(u˜ + ε, u) ↘ u on Ω as ε → 0, therefore (ddc max(u˜ + ε, u))m ∧ βn−m is weakly convergent to (ddcu)m ∧ βn−m as ε → 0. On the other hand, 1K is upper semicontinuous on Ω so we can approximate 1K with a decreasing sequence of continuous functions ϕj . So, lim sup ε→0  Ω 1K(dd c max(u˜+ ε, u))m ∧ βn−m = lim sup ε→0 [ lim j  Ω ϕj(dd c max(u˜+ ε, u))m ∧ βn−m ] ≤ lim sup ε→0 (  Ω ϕj(dd c max(u˜+ ε, u))m ∧ βn−m ) ≤  Ω ϕj(dd cu)m ∧ βn−m ↘  K (ddcu)m ∧ βn−m as j →∞. This yields (ddcu˜)m ∧ βn−m ≤ (ddcu)m ∧ βn−m on {u˜ = u} ∩Ω. 152 H. M. Le et al. Step 3. Now, let E0m(Ω) ∩ C(Ω) 3 uj ↘ u be such that sup j  Ω (ddcuj) m ∧ βn−m <∞. By Step 2, we have  Ω˜ (ddcu˜j) m ∧ βn−m =  {u˜j=uj}∩Ω (ddcu˜j) m ∧ βn−m ≤  {u˜j=uj}∩Ω (ddcuj) m ∧ βn−m ≤  Ω (ddcuj) m ∧ βn−m. Hence, sup j  Ω˜ (ddcu˜j) m ∧ βn−m ≤ sup j  Ω (ddcuj) m ∧ βn−m <∞. Thus, u˜ := limj→∞ u˜j ∈ Fm(Ω˜) and u˜ ≤ u on Ω. The following result deals with the locality of membership in Em(Ω). Theorem 3.3. Let Ω be a bounded hyperconvex domain in Cn and m be an integer with 1 ≤ m ≤ n. Assume that u ∈ SH−m(Ω). Then the following statements are equivalent: (i) u ∈ Em(Ω). (ii) For all K b Ω, there exists a sequence {uj} ⊂ E0m(Ω) ∩ C(Ω), uj ↘ u on K, such that for all p = 0, 1, . . . ,m we have sup j  K (−uj)p(ddcuj)m−p ∧ βn−m+p <∞. (iii) For every W b Ω such that W is a hyperconvex domain, we have u|W ∈ Em(W ). (iv) For every z ∈ Ω there exists a hyperconvex domain Vz b Ω such that z ∈ Vz and u|Vz ∈ Em(Vz). Proof. The proof here is due to Błocki [B2]. (i)⇒(ii). Let K b Ω be given. Since u ∈ Em(Ω), there exists v ∈ Fm(Ω) with v = u onK. By the definition of the class Fm(Ω) there exists a sequence {uj} ⊂ E0m(Ω) ∩ C(Ω) with uj ↘ v on Ω such that (3.1) sup j  Ω (ddcuj) m ∧ βn−m <∞. Then uj ↘ u on K. We have to prove sup j  K (−uj)p(ddcuj)m−p ∧ βn−m+p <∞ Some characterizations of the class Em(Ω) 153 for p = 0, 1, . . . ,m. It is obvious that the conclusion holds for p = 0. Assume that 1 ≤ p ≤ m. Choose R > 0 such that ‖z‖2 − R2 < 0 on Ω and assume that ϕ ∈ E0m(Ω) is given. Next, we choose A > 0 such that ‖z‖2 −R2 ≥ Aϕ on K. Set h = max(‖z‖2 − R2, Aϕ). Then h ∈ E0m(Ω) and ddch = β on K. For each p = 1, . . . ,m we define Ip =  Ω (−uj)p(ddcuj)m−p ∧ (ddch)p ∧ βn−m. Then by integration by parts we get the chain of inequalities  K (−uj)p(ddcuj)m−p ∧ βn−m+p =  K (−uj)p(ddcuj)m−p ∧ (ddch)p ∧ βn−m ≤ Ip =  Ω h(ddcuj) m−p ∧ (ddch)p−1 ∧ (ddc(−uj)p) ∧ βn−m =  Ω h(ddcuj) m−p ∧ (ddch)p−1[p(p−1)duj ∧ dcuj−p(−uj)p−1ddcuj ]∧βn−m ≤ p  Ω (−h)(−uj)p−1(ddcuj)m−p+1 ∧ (ddch)p−1 ∧ βn−m ≤ p‖h‖Ω  Ω (−uj)p−1(ddcuj)m−p+1 ∧ (ddch)p−1 ∧ βn−m = p‖h‖ΩIp−1 ≤ · · · ≤ p!‖h‖pΩI0 = p!‖h‖pΩ  Ω (ddcuj) m ∧ βn−m. Hence, by (3.1) we get sup j  K (−uj)p(ddcuj)m−p ∧ βn−m+p ≤ p!‖h‖pΩ sup j  Ω (ddcuj) m ∧ βn−m <∞. (ii)⇒(iii). Let W b Ω be a hyperconvex domain. Take U bW b Ω and a sequence E0m(Ω) 3 uj ↘ u on W such that sup j  W (−uj)p(ddcuj)m−p ∧ βn−m+p <∞ for p = 0, 1, . . . ,m. Set u˜j = sup{ϕ ∈ SH−m(W ) : ϕ ≤ uj on U} ∈ E0m(W ). Next, choose U b Ω1 b · · · b Ωm b W . Since uj ≤ u˜j on W and (ddcu˜j)m ∧ βn−m = 0 on W\U , by applying Lemma 3.1 repeatedly we arrive at  W (ddcu˜j) m ∧ βn−m =  U (ddcu˜j) m ∧ βn−m .  Ω1 (ddcuj + (−uj)β) ∧ (ddcu˜j)m−1 ∧ βn−m 154 H. M. Le et al. =  Ω1 (ddcu˜j) m−1 ∧ ddcuj ∧ βn−m +  Ω1 (−uj)(ddcu˜j)m−1 ∧ βn−m+1 .  Ω2 (ddcuj + (−uj)β) ∧ (ddcu˜j)m−2 ∧ ddcuj ∧ βn−m +  Ω2 (−uj)(ddcuj + (−uj)β) ∧ (ddcu˜j)m−2 ∧ βn−m+1 .  Ω2 [(−uj)2β2 + (−uj)β ∧ ddcuj + (ddcuj)2] ∧ (ddcu˜j)m−2 ∧ βn−m . · · · .  Ωm [(−uj)mβm + (−uj)m−1ddcuj ∧ βm−1 + · · ·+ (ddcuj)m] ∧ βn−m. Hence, sup j  W (ddcu˜j) m ∧ βn−m . sup j  Ωm [(−uj)mβm + (−uj)m−1ddcuj ∧ βm−1 + · · ·+ (ddcuj)m] ∧ βn−m . sup j  W [(−uj)mβm+(−uj)m−1ddcuj ∧βm−1+ · · ·+(ddcuj)m]∧βn−m<∞. Thus, uU,W := lim u˜j ∈ Fm(W ). Since U b W is arbitrary and uU,W = u on U , we conclude that u ∈ Em(W ). (iii)⇒(iv). This is obvious. (iv)⇒(i). Assume that Ω′ b Ω. Choose zj ∈ Ω, j = 1, . . . , s, such that Ω′ b ⋃s j=1 Vzj , where Vzj are hyperconvex domains. Let Wzj b Vzj be such that Ω′ b ⋃s j=1Wzj . Since u|Vzj ∈ Em(Vzj ), there exists vj ∈ Fm(Vzj ) such that vj = u onWzj . By Lemma 3.2 there exists v˜j ∈ Fm(Ω) such that v˜j ≤ vj on Vzj . The convexity of the class Fm(Ω) implies that v˜ := 1s v˜1 + · · ·+ 1s v˜s ∈Fm(Ω), and hence max(v˜, u) ∈ Fm(Ω). However, max(v˜, u) = u on Ω′ and therefore u ∈ Em(Ω). The proof of Theorem 3.3 is complete. Remark 3.4. In [B2], Błocki proved that membership in the class E(Ω) is a local property. From Theorem 3.3 we get the following property of Em(Ω). Corollary 3.5. Assume that Ω is a bounded hyperconvex domain. Then Em(Ω) ⊂ Em−1(Ω). Proof. Assume that u ∈ Em(Ω). Let K b Ω. Take a domain Ω′ with K b Ω′ b Ω. By Theorem 3.3 there exists a sequence {uj} ⊂ E0m(Ω) such Some characterizations of the class Em(Ω) 155 that uj ↘ u on Ω′ and sup j  Ω′ (−uj)p(ddcuj)m−p ∧ βn−m+p <∞ for p = 0, 1, . . . ,m. Choose h ∈ E0m−1(Ω). For each j > 0 take mj > 0 such that uj ≥ mjh on Ω′. Set vj = max(uj ,mjh) ∈ E0m−1(Ω) and vj = uj on Ω′. Note that vj ↘ u on Ω′ and (ddcvj)p ∧ βq = (ddcuj)p ∧ βq on Ω′ for 0 ≤ p ≤ m − 1 and 1 ≤ q ≤ n −m + 1. We may assume that u|Ω′ ≤ −1. By Hartogs’ lemma (see [Ho, Theorem 3.2.13]) we conclude that vj |Ω′ ≤ −1 for j ≥ j0 with some j0. Without loss of generality, we